- .stream() and .astream() call on_llm_new_token, removing the need for
subclasses to do so. Backwards compatible because now we don't pass
run_manager into ._stream and ._astream
- .generate() and .agenerate() now handle `stream: bool` kwarg for
_generate and _agenerate. Subclasses handle this arg by delegating to
._stream(), now one less thing they need to do. Backwards compat because
this is an optional arg that we now never pass to the subclasses
- .generate() and .agenerate() now inspect callback handlers to decide
on a default value for stream:bool if not passed in. This auto enables
streaming when using astream_events and astream_log
- as a result of these three changes any usage of .astream_events and
.astream_log should now yield chat model stream events
- In future PRs we can update all subclasses to reflect these two things
now handled by base class, but in meantime all will continue to work
**Description:**
This PR adds a slightly more helpful message to a Tool Exception
```
# current state
langchain_core.tools.ToolException: Too many arguments to single-input tool
# proposed state
langchain_core.tools.ToolException: Too many arguments to single-input tool. Consider using a StructuredTool instead.
```
**Issue:** Somewhat discussed here 👉#6197
**Dependencies:** None
**Twitter handle:** N/A
---------
Co-authored-by: Bagatur <22008038+baskaryan@users.noreply.github.com>
As mentioned in #18322, the current PydanticOutputParser won't work for
anyone trying to parse to pydantic v2 models. This PR adds a separate
`PydanticV2OutputParser`, as well as a `langchain_core.pydantic_v2`
namespace that will fail on import to any projects using pydantic<2.
Happy to update the docs for output parsers if this is something we're
interesting in adding.
On a separate note, I also updated `check_pydantic.sh` to detect
pydantic imports with leading whitespace and excluded the internal
namespaces. That change can be separated into its own PR if needed.
---------
Co-authored-by: Jan Nissen <jan23@gmail.com>
Patch potential XML vulnerability CVE-2024-1455
This patches a potential XML vulnerability in the XMLOutputParser in
langchain-core. The vulnerability in some situations could lead to a
denial of service attack.
At risk are users that:
1) Running older distributions of python that have older version of
libexpat
2) Are using XMLOutputParser with an agent
3) Accept inputs from untrusted sources with this agent (e.g., endpoint
on the web that allows an untrusted user to interact wiith the parser)
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: If you're adding a new integration, please
include
1. a test for the integration, preferably unit tests that do not rely on
network access,
2. an example notebook showing its use. It lives in
`docs/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, hwchase17.
DefusedXML is causing parsing errors on previously functional code with
the 0.7.x versions. These do not seem to support newer version of python
well. 0.8.x has only been released as rc, so we're not going to to use
it in the core package
Few-Shot prompt template may use a `SemanticSimilarityExampleSelector`
that in turn uses a `VectorStore` that does I/O operations.
So to work correctly on the event loop, we need:
* async methods for the `VectorStore` (OK)
* async methods for the `SemanticSimilarityExampleSelector` (this PR)
* async methods for `BasePromptTemplate` and `BaseChatPromptTemplate`
(future work)
Previous PR passed _parser attribute which apparently is not meant to be
used by user code and causes non deterministic failures on CI when
testing the transform and a transform methods. Reverting this change
temporarily.
This mitigates a security concern for users still using older versions of libexpat that causes an attacker to compromise the availability of the system if an attacker manages to surface malicious payload to this XMLParser.
For prompt templates with only 1 variable (common in e.g.,
MessageGraph), it's convenient to wrap the incoming object in the
variable before formatting.
The downside of this, of course, would be that some number of
invocations will successfully format when the user may have intended to
format it properly before
Classes and functions defined in __init__.py are not parsed into the API
Reference.
For example:
- libs/core/langchain_core/messages/__init__.py : AnyMessage,
MessageLikeRepresentation, get_buffer_string(), messages_from_dict(),
...
Opinionated: __init__.py is not a typical place to define artifacts.
Moved artifacts from __init__ into utils.py.
Added `MessageLikeRepresentation` to __all__ since it is used outside of
`messages`, for example, in
`libs/core/langchain_core/language_models/base.py`
Added `_message_from_dict` to __all__ since it is used outside of
`messages`(???) I would add `message_from_dict` (without underscore) as
an alias. Please, advise.
Covered by tests in
`libs/core/tests/unit_tests/language_models/chat_models/test_base.py`,
`libs/core/tests/unit_tests/language_models/llms/test_base.py` and
`libs/core/tests/unit_tests/runnables/test_runnable_events.py`
**Description:**
Currently, `CacheBackedEmbeddings` computes vectors for *all* uncached
documents before updating the store. This pull request updates the
embedding computation loop to compute embeddings in batches, updating
the store after each batch.
I noticed this when I tried `CacheBackedEmbeddings` on our 30k document
set and the cache directory hadn't appeared on disk after 30 minutes.
The motivation is to minimize compute/data loss when problems occur:
* If there is a transient embedding failure (e.g. a network outage at
the embedding endpoint triggers an exception), at least the completed
vectors are written to the store instead of being discarded.
* If there is an issue with the store (e.g. no write permissions), the
condition is detected early without computing (and discarding!) all the
vectors.
**Issue:**
Implements enhancement #18026.
**Testing:**
I was unable to run unit tests; details in [this
post](https://github.com/langchain-ai/langchain/discussions/15019#discussioncomment-8576684).
---------
Signed-off-by: chrispy <chrispy@synopsys.com>
Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>
Classes and functions defined in __init__.py are not parsed into the API
Reference.
For example: libs/core/langchain_core/globals/__init__.py :
`set_verbose` `get_llm_cache`, `set_llm_cache`, ...
And the whole `langchain_core.globals` namespace is not visible in the
API Reference. The refactoring is just file renaming.
- **Description:** Enhanced the `BaseChatModel` to support an
`Optional[Union[bool, BaseCache]]` type for the `cache` attribute,
allowing for both boolean flags and custom cache implementations.
Implemented logic within chat model methods to utilize the provided
custom cache implementation effectively. This change aims to provide
more flexibility in caching strategies for chat models.
- **Issue:** Implements enhancement request #17242.
- **Dependencies:** No additional dependencies required for this change.
---------
Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>
Issue : For functions which have an argument with the name 'title', the
convert_pydantic_to_openai_function generates an incorrect output and
omits the argument all together. This is because the _rm_titles function
removes all instances of the the key 'title' from the output.
Description : Updates the _rm_titles function to check the presence of
the 'type' key as well before removing the 'title' key. As the title key
that we wish to omit always has a type key along with it.
Potential gap if there is a function defined which has both title and
key as argument names, in which case this would fail. Maybe we could set
a filter on the function argument names and reject those with keyword
argument names.
No dependencies. Passed all tests.
- [x] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [x] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [x] **Add tests and docs**: If you're adding a new integration, please
include
1. a test for the integration, preferably unit tests that do not rely on
network access,
2. an example notebook showing its use. It lives in
`docs/docs/integrations` directory.
- [x] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, hwchase17.
---------
Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>
The root run id (~trace id's) is useful for assigning feedback, but the
current recommended approach is to use callbacks to retrieve it, which
has some drawbacks:
1. Doesn't work for streaming until after the first event
2. Doesn't let you call other endpoints with the same trace ID in
parallel (since you have to wait until the call is completed/started to
use
This PR lets you provide = "run_id" in the runnable config.
Couple considerations:
1. For batch calls, we split the trace up into separate trees (to permit
better rendering). We keep the provided run ID for the first one and
generate a unique one for other elements of the batch.
2. For nested calls, the provided ID is ONLY used on the top root/trace.
```
chain.invoke("foo", {"run_id": uuid.uuid4()})
```
- Description: [a description of the change] Add in code documentation
to core Runnable with_fallbacks method (docs only)
- Issue: the issue #18804
@eyurtsev PTAL
- **Description:** Handling fallbacks when calling async streaming for a
LLM that doesn't support it.
- **Issue:** #18920
- **Twitter handle:**@maximeperrin_
---------
Co-authored-by: Maxime Perrin <mperrin@doing.fr>
poetry can't reliably handle resolving the number of optional "extended
test" dependencies we have. If we instead just rely on pip to install
extended test deps in CI, this isn't an issue.
**PR message**: ***Delete this entire checklist*** and replace with
- **Description:** [a description of the change](docs: Add in code
documentation to core Runnable assign method)
- **Issue:** the issue #18804
